Browsing by Author "Kämäräinen, Matti"

The 2-meter temperature output (daily mean, minimum and maximum values) of a six-member regional climate model ensemble and the corresponding observations for three stations in Finland (Helsinki, Jyväskylä and Sodankylä) are used to produce future temperature projections. Both the observed (‘delta change’ -approach) and the model scenario ('bias correction' - approach) data series are statistically corrected with several different methods. These methods make use of the statistics of temperature between the 30-year periods of observations, model control and model scenario simulations, and vary from simple (adjusting of mean) to complex (quantile mapping). Each month is processed separately. The main projection experiments are I) from 1951-1980 to 1981-2010 and II) from 1981-2010 to 2011-2040, 2041-2070 and 2069-2098. The method-dependent and to a lesser extent the model-dependent results are evaluated by means of root mean square error, mean error (mean bias), the location of quantile points, the number of daily frequency indices, analysis of variance and sensitivity tests. In near-term projections (e.g. from 1981-2010 to 2011-2040) the more conservative delta change methods slightly outperform the bias correction methods. In mid-term (projections to 2041-2070) and especially in far-term (projections to 2069-2098) predictions the bias correction approach is better in cross validation. The complicated shape of winter-time temperature distributions emphasizes the importance of correct handling of the biases compared to southern, less snowy areas. For that reason the detailed quantile mapping type bias correction approach produces the best results with predictions scoping to the end of the century.